About FraudMap
Nick Shirley travels the country filming investigations and on-the-ground reports, and the information in those videos — places, businesses, dollar figures — was scattered across hours of footage. I was curious what it all looked like as a dataset: every location, every named entity, every number he's reported, across the entire catalog. So I built a pipeline that turns each video into structured data and put the whole thing on one map — with every figure linked to the exact second in the video where it's said.
What this is (and isn't)
FraudMap is a free, independent data project. It is not affiliated with Nick Shirley — I analyze what he publishes, the same as any viewer, and index it. Nothing here is original reporting; every figure is a claim made in a video, with a citation link so you can hear it in context.
Nothing here is confirmed or validated by this site. The pipeline quantifies and categorizes what is said in the videos — it checks extracted figures against the transcript, never against the underlying facts. Most figures are allegations, not adjudicated findings: every dollar amount is labeled with the status stated in the video — alleged, under investigation, charged, convicted, or an official estimate — and program budgets or industry sizes are never blended into fraud totals.
Every person and business named is presumed innocent; appearing here does not imply wrongdoing. Full details are on the disclaimer page.
How it works
- An automated pipeline checks the channel every 6 hours for new videos and pulls each video's auto-generated transcript.
- Claude (Anthropic's Opus model) reads the transcript and extracts structured data: filming locations, businesses and programs named, and every dollar figure — with who said it, what it covers, and the timestamp where it appears.
- A second, independent AI pass acts as an adversarial fact-checker: it re-reads the transcript trying to refute each extracted claim. Anything it can't support gets rejected and never reaches the site; corrections are stored and shown.
- Totals follow strict double-counting rules: money tied to a specific business adds up, scheme-level estimates ("billions in Minnesota") are counted once per scheme, and when the same business's figure appears in multiple videos it is counted once — in the earliest video that reported it.
Every number on this site can be traced: claim → quote → ▶ timestamp → video. If you find an error, that trail is how you check it — and the transparency page shows exactly what's been ingested, when, and what failed.
Known limits
Transcripts can't see documents shown on camera, auto-captions sometimes garble names (reconstructed names are flagged with ~), and brand-new uploads occasionally wait on captions before they can be processed. When a name is garbled beyond confident reconstruction, it stays out rather than being guessed.
Who built this
FraudMap was built by Oze Botach — @obotach — entrepreneur, engineer, architect — and ENDUURE, his software agency, which engineers the systems businesses run on — custom platforms, ERPs, AI-integrated systems, and automation. Infrastructure over hype.
This whole site — the ingestion pipeline, the AI extraction and verification, the map — is the kind of system ENDUURE builds for clients. This one is free because the data should be public.
Contact
Corrections, ideas, or press: DM @obotach on X